Machine learning device, insulation state diagnosis device, machine learning method, machine learning program, insulation state diagnosis method, and insulation state diagnosis program

Abstract

The invention provides a machine learning device capable of diagnosing the insulation state of a winding without stopping a shield motor pump. A machine learning device (4) generates a learning model (6) for an insulation state diagnosis device (5) for diagnosing the insulation state of a winding (31) of a stator (23) used in a shield motor pump (2). A machine learning device (4) uses unprocessed data or data obtained by pre-processing at least a portion of the unprocessed data as input data, and inputs learning data including at least the input data into a learning model (6). Therefore, the learning model learns the correlation between the input data and the diagnosis information of the insulation state of the winding. The unprocessed data is data of a voltage value and a current value supplied to the motor unit (20), data of a temperature of a winding (31) of the stator (23) or a temperature, a temperature difference, or a pressure related thereto, and data of an inflow temperature and a flow rate of the working fluid in the pump unit (10).

Description

Technical Field

[0001] The present invention relates to a machine learning apparatus, an insulation condition diagnosis apparatus, a machine learning method, a machine learning program, an insulation condition diagnosis method, and an insulation condition diagnosis program suitable for diagnosing the insulation condition of the windings of a shielded electric motor pump having an electric motor section and a pump section. Background Technology

[0002] In a canned motor pump, the rotor of the motor that drives the pump unit is housed within a casing filled with working fluid. The stator of the motor unit is housed in a motor housing located on the outer periphery of the casing and functioning as a frame.

[0003] In this type of canned motor pump, the motor output is set based on the allowable temperature of the stator coil. That is, because the rotor of the canned motor pump is surrounded by working fluid, it can be cooled rapidly. Therefore, the heat-generating component of the canned motor pump is mainly on the stator side, with the highest temperature at the coil ends. Therefore, in the prior art, the coil end temperature is ensured not to exceed the allowable temperature by specifying the motor output and the fluid temperature.

[0004] In addition, under normal circumstances, a canned motor pump will introduce a portion of the working fluid into the housing containing the rotor, which will lubricate and cool the bearings, absorb the heat from the motor section (rotor, etc.), and then return it to the pump section.

[0005] However, canned motor pumps are commonly used in factories handling high-temperature working fluids. Therefore, when the high-temperature working fluid is introduced into the casing, the motor windings are at a temperature higher than the working fluid, easily exposing them to a high-temperature environment exceeding the permissible range of heat-resistant insulation. Consequently, prolonged use of the motor section of a canned motor pump in high-temperature environments carries a risk of winding insulation degradation.

[0006] When the insulation of the windings deteriorates significantly, the windings need to be replaced. However, since the stator is sealed between the housing and the motor casing, replacing the stator windings requires disassembling the sealed motor section. Such disassembly is extremely costly and time-consuming. Therefore, it is necessary to properly assess the winding life, i.e., the insulation condition of the windings.

[0007] Therefore, in the prior art, in order to perform winding deterioration diagnosis, the shielded motor pump is stopped, the insulation resistance and terminal resistance of the winding are measured periodically and compared with normal values ​​or historical measurement values, thereby managing the trend of insulation status (see Patent Document 1).

[0008] Existing technical documents

[0009] Patent documents

[0010] Patent Document 1: Japanese Patent Application Publication No. 2010-256348 Summary of the Invention

[0011] The problem that the invention aims to solve

[0012] However, in order to manage the trend of the insulation condition of the winding, it is necessary to stop the canned motor pump to measure the resistance value (insulation resistance, inter-terminal resistance). In factory operation, it is difficult to stop the canned motor pump frequently.

[0013] The present invention was made in view of the above-mentioned problems, and its main objective is to provide a machine learning device, an insulation condition diagnosis device, a machine learning method, a machine learning program, an insulation condition diagnosis method, and an insulation condition diagnosis program that can diagnose the insulation condition of windings without stopping the shielded motor pump.

[0014] Technical solutions for solving the problem

[0015] One embodiment of the present invention involves a machine learning apparatus that generates a learning model for an insulation condition diagnostic device used to diagnose the insulation condition of the stator windings in a shielded electric motor pump. The shielded electric motor pump has an electric motor section and a pump section driven by the electric motor section. The motor unit includes: a rotor capable of rotating about an axis; a stator opposite the rotor with a gap; and a cylindrical housing having an inner rotor chamber for accommodating the rotor and an outer stator chamber for accommodating the stator together with the motor housing. The shielded motor pump circulates a portion of the working fluid in the pump section to the rotor chamber, wherein... The machine learning device has: The learning data acquisition unit takes unprocessed data or data that has undergone preprocessing on at least a portion of the unprocessed data as input data, and acquires a plurality of sets of learning data having at least the input data. The unprocessed data includes data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator windings or related temperatures, temperature differences or pressures, and data on the inflow temperature and flow rate of the working fluid. The machine learning unit, by inputting the learning data into the learning model, enables the learning model to learn the correlation between the input data and diagnostic information regarding the insulation state of the winding; and The learned model storage unit stores the learned models learned by the machine learning unit.

[0016] Furthermore, one embodiment of the insulation condition diagnostic device according to the present invention is used to diagnose the insulation condition of the stator windings used in a shielded electric motor pump using a learning model generated by the aforementioned machine learning device. The insulation condition diagnostic device has the following features: The input data acquisition unit acquires input data that includes unprocessed data or data for which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor unit, temperature of the stator windings or related temperatures, temperature differences, or pressures, and inflow temperature and flow rate of the working fluid. The inference unit inputs the input data acquired by the input data acquisition unit into the learning model to infer diagnostic information about the insulation state of the stator winding.

[0017] Furthermore, one embodiment of the machine learning method involved in this invention is used to learn a learning model, which is used in an insulation condition diagnostic device to diagnose the insulation condition of the stator windings used in a shielded electric motor pump. The shielded electric motor pump has an electric motor section and a pump section driven by the electric motor section. The motor unit includes: a rotor capable of rotating about an axis; a stator opposite the rotor with a gap; and a cylindrical housing having an inner rotor chamber for accommodating the rotor and an outer stator chamber for accommodating the stator together with the motor housing. The shielded motor pump circulates a portion of the working fluid in the pump section to the rotor chamber, wherein... The machine learning method has the following characteristics: The learning data acquisition process takes unprocessed data or data in which at least a portion of the unprocessed data has been preprocessed as input data, and acquires a plurality of sets of learning data having at least the input data. The unprocessed data includes data on voltage and current values ​​supplied to the motor section, data on the temperature of the stator windings or related temperatures, temperature differences or pressures, and data on the inflow temperature and flow rate of the working fluid. The machine learning process involves inputting the learning data into the learning model, thereby enabling the learning model to learn the correlation between the input data and diagnostic information regarding the insulation state of the winding; and The learned model storage process stores the learned model learned by the machine learning process into the learned model storage unit.

[0018] Furthermore, one embodiment of the machine learning program involved in this invention enables a computer to perform the various steps of the machine learning method.

[0019] Furthermore, one embodiment of the insulation condition diagnosis method of the present invention uses a learning model generated by the machine learning device to diagnose the insulation condition of the stator windings used in the shielded motor pump, wherein... The insulation condition diagnosis method has the following characteristics: The input data acquisition process acquires input data that includes unprocessed data or data for which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor unit, temperature of the stator windings or related temperatures, temperature differences, or pressures, and inflow temperature and flow rate of the working fluid. The inference process involves inputting the input data obtained from the input data acquisition process into the learning model to infer diagnostic information about the insulation state of the stator windings.

[0020] Furthermore, the insulation condition diagnosis program of one embodiment of the present invention enables a computer to execute the various steps of the insulation condition diagnosis method.

[0021] Invention Effects

[0022] This invention provides a machine learning device, an insulation condition diagnosis device, a machine learning method, a machine learning program, an insulation condition diagnosis method, and an insulation condition diagnosis program that can diagnose the insulation condition of windings without stopping the shielded motor pump. Attached Figure Description

[0023] Figure 1 This is a schematic structural diagram of an insulation condition diagnostic device applied to a shielded motor pump.

[0024] Figure 2 This is a cross-sectional view showing an example of a canned motor pump.

[0025] Figure 3 This is a block diagram representing an example of a machine learning device.

[0026] Figure 4 This is a flowchart illustrating an example of a machine learning method using a machine learning apparatus according to the first embodiment.

[0027] Figure 5 This is a block diagram representing an example of an insulation condition diagnostic device.

[0028] Figure 6 This is a flowchart illustrating an example of an insulation condition diagnosis method for an insulation condition diagnosis device according to the first embodiment.

[0029] Figure 7 This is a cross-sectional view showing other examples of canned motor pumps.

[0030] Figure 8A It means applied to Figure 7 A block diagram of an example of a machine learning device for a shielded electric motor pump is shown.

[0031] Figure 8B It means applied to Figure 7 A block diagram illustrating an example of an insulation condition diagnostic device for a shielded motor pump.

[0032] Figure 9 This is a flowchart illustrating an example of a machine learning method using a machine learning apparatus according to the second embodiment.

[0033] Figure 10 This is a flowchart illustrating an example of an insulation condition diagnosis method for an insulation condition diagnosis device according to the second embodiment. Detailed Implementation

[0034] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0035] (First Implementation)

[0036] Figure 1 This is a schematic structural diagram of an insulation condition diagnostic system 1 applied to a canned motor pump. The canned motor pump 2 used herein can be exemplified as a horizontally mounted canned motor pump with the motor axis direction approximately horizontal, but the invention is not limited to the horizontal type; it can also be a vertically mounted canned motor pump with the motor axis direction approximately vertical.

[0037] First, use Figure 2 This section provides an overview of the canned motor pump 2.

[0038] Figure 2 This is a cross-sectional view showing a structural example of the canned motor pump 2 according to this embodiment. Furthermore, the X direction described in the figure is the direction of the motor axis, the right side of the figure is the front, and the left side of the figure is the rear.

[0039] exist Figure 2 In the process, the shielded motor pump 2 has a pump section 10 with the structure of a centrifugal pump and a motor section 20 that drives the pump section 10. The pump section 10 and the motor section 20 are coaxially connected via the shaft 21 of the motor 24, which will be described later.

[0040] The pump section 10 includes an impeller 11 and a pump housing 13 forming a pump chamber 12 that accommodates the impeller 11. The impeller 11 is coupled to the front end of the shaft 21 of the motor 24. Furthermore, in this embodiment, the pump section 10 is described as being constructed as a centrifugal pump, but the present invention is not limited thereto, and the pump section 10 may also be constructed as a mixed-flow pump, an axial-flow pump, or a turbopump.

[0041] The motor section 20 has a motor 24 and a motor housing 25 that houses the motor 24. The motor 24 includes a rotor 22 that is fixed around the shaft 21 and rotates integrally with the shaft, and an annular stator 23 that is arranged relative to the rotor 22 at predetermined intervals.

[0042] The stator 23 has a stator core 30 with teeth arranged circumferentially on the inner circumferential surface of a generally cylindrical shape, and a winding (coil) 31 formed by winding a conductor around the teeth of the stator core 30. The ends of the conductors of the winding 31 are guided into a terminal box 32 that extends radially outward from the outer cylinder 41 of the motor housing 25 and are connected to terminals (not shown) provided in the terminal box 32.

[0043] A cylindrical housing 33 is disposed on the inner side of the stator core 30, i.e., on the outer side of the rotor 22, that is, between the rotor 22 and the stator 23. The housing 33 is fixed in contact with the inner circumferential surface of the stator core 30, i.e., the tip surface of the teeth, and is disposed at a certain interval from the outer circumferential surface of the rotor 22.

[0044] Additionally, a cylindrical outer cylinder (stator belt) 41 for accommodating the motor is provided along the outer circumferential surface of the stator core 30. The housing 33 and the outer cylinder 41 are concentrically arranged and are formed to have approximately the same length as the axial direction X. Furthermore, at each end of the housing 33 and the outer cylinder 41, annular end plates 42 and 43 are provided to block the cylindrical space formed between the housing 33 and the outer cylinder 41.

[0045] The end plate 42 on the pump section 10 side is bolted to the pump housing 13. Additionally, the end plate 43 on the opposite side of the pump section 10 is bolted to the bearing cage 44 that blocks the space inside the housing 33.

[0046] The outer cylinder 41, the end plates 42 and 43 at both ends, and the bearing cage 44 together form the motor housing 25 that houses the motor 24.

[0047] In addition, a rotor chamber 50 for accommodating the rotor 22 is formed inside the housing 33 via the housing 33, the pump section 10, and the bearing cage 44.

[0048] Furthermore, a stator chamber 45 for accommodating the stator 23 is formed by a cylindrical space divided by the outer cylinder 41, the housing 33, and the end plates 42 and 43 at both ends. The stator chamber 45 has a stator core 30 arranged in the approximately central part along the axial direction, and the stator core 30 has annular coil end receiving portions 45a and 45b in front and behind it along the axial direction to accommodate the coil ends 31a and 31b.

[0049] In addition, cylindrical support sleeves 34 (34a, 34b) for preventing deformation of the housing 33 are tightly attached to the outer peripheral surfaces of the coil end receiving portions 45a, 45b of the housing 33.

[0050] The shaft 21, which is integrally fixed to the rotor 22, is supported at both ends by bearings 51 and 52. The bearing 51 on the pump section 10 side is held on a bearing portion 53 extending from a bearing cage 44 fixed to the end plate 42 behind the pump housing 13 of the pump section 10. Conversely, the bearing 52 on the opposite side of the pump section 10 is held on a bearing portion 54 extending from a bearing cage 46 fixed to the end plate 43. These bearings 51 and 52 are composed of planar bearings positioned in the bearing portions 53 and 54.

[0051] Furthermore, the aforementioned shielded electric motor pump 2 is fixed to the support platform 56 via the feet 55 fixed to the end plates 42 and 43.

[0052] The bearing cage 44, which is fixed to the back of the pump housing 13 of the pump section 10, forms an inlet flow path 14 for guiding the working fluid within the housing 33. Additionally, a central hole 15 is formed along the entire length of the shaft 21. A portion of the working fluid delivered through the impeller 11 is as follows... Figure 2 As indicated by the arrow, the internal circulating fluid circulates within the housing 33. Specifically, it spreads towards the rear of the impeller 11 and is introduced into the housing 33 (rotor chamber 50) via the inlet flow path 14. After the working fluid introduced into the housing 33 (rotor chamber 50) lubricates and cools the bearings 51 and 52, and cools the motor section 20 (rotor 22, etc.), it enters the central hole 15 formed on the shaft from the rear end of the shaft 21. Then, the working fluid returns to the suction side of the pump section 10 via the central hole 15 and is discharged into the vortex chamber through the impeller 11.

[0053] For the shielded electric motor pump 2 with the above structure, an insulation condition diagnosis system 1 is provided to diagnose the insulation condition of the winding 31 of the stator 23. For example... Figure 1 As shown, the insulation condition diagnostic system 1 includes a measuring device 3, a machine learning device 4, and an insulation condition diagnostic device 5.

[0054] The measuring device 3 is installed in a suitable location near or around the shielded motor pump 2 to measure the physical and state quantities of each part. During the learning phase, the measuring device 3 is used to connect with the machine learning device 4 to measure data for learning, and during the inference phase, it is used to connect with the insulation condition diagnostic device 5 to measure data for diagnostic purposes.

[0055] The measuring device 3 used in the learning phase can be the same as or different from the measuring device used in the inference phase. It includes a motor voltage and current sensor 101, a winding heating state measuring sensor 102, and a working fluid measuring sensor 103.

[0056] The motor voltage and current sensor 101 measures the motor voltage and current values ​​supplied to the motor unit 20. Therefore, for example, it can be installed on the power line connecting the motor 24 and the motor control device 26 to measure the voltage and current values ​​supplied from the motor control device 26 to the motor 24. Alternatively, the motor voltage and current sensor 101 can be installed inside the motor unit 20 or inside the motor control device 26.

[0057] In addition, the motor voltage and current sensor 101 can also measure the voltage and current values ​​of each of the three phases.

[0058] The winding heating condition measuring sensor 102 directly or indirectly measures the temperature of the winding 31 of the motor 24 (stator 23). That is, the winding heating condition measuring sensor 102 measures the temperature of the winding or related temperature, temperature difference or pressure data.

[0059] When directly measuring the temperature of the winding 31, thermocouples can be installed at the locations of the highest temperature in the winding 31, such as the coil ends 31a and 31b, to measure the temperature. Alternatively, thermocouples can be installed inside the winding 31 within the stator core 30 to measure the winding temperature.

[0060] In the case of indirectly measuring the temperature of winding 31, thermocouples can be installed on the support sleeves 34a, 34b, end plates 42, 43, and bearing cages 44, 46, which are affected by the radiant heat from winding 31 and the state (amount, temperature) of the internal circulating fluid, to measure the surface temperature of the part.

[0061] Among them, the example of using a thermocouple to detect temperature is illustrated in the winding heating state measuring sensor 102. However, other contact sensors such as thermistors and temperature-sensing resistors can also be used, as well as non-contact sensors such as infrared sensors.

[0062] In addition, in order to indirectly measure the temperature of winding 31, the pressure P1 of stator chamber 45 can be measured.

[0063] The working fluid measuring sensor 103 is composed, for example, an inflow temperature sensor 103a and a flow sensor 103b disposed in an inflow pipe 60 connected to the inflow inlet 13a of the pump housing 13.

[0064] The inflow temperature sensor 103a measures the temperature of the working fluid flowing into the pump section 10. Therefore, for example, the fluid temperature in the inflow pipe 60 can be measured by using a piping temperature sensor that uses a temperature-sensing resistor as a temperature detection element. In addition, the flow sensor 103b measures the inflow rate of the working fluid into the pump section 10. Therefore, various sensors such as ultrasonic, electromagnetic, Karman vortex, impeller, float, thermal, and differential pressure sensors can be used.

[0065] The measuring device 3 is configured to measure the physical quantities and state quantities of each part, and input the measured values ​​to the machine learning device 4 or the insulation state diagnostic device 5. The measured values ​​measured by the measuring device 3 can be precise measured values ​​measured at a specified time, discrete sets of measured values ​​over time measured in each specified measurement cycle within a specified period, or continuous measured values ​​measured within a specified period.

[0066] The machine learning device 4 acts as the main body of the learning phase, generating the learning model 6 used to diagnose the insulation state of the winding 31 through machine learning. As for machine learning, either "supervised learning" or "unsupervised learning" can be used. "Supervised learning" is used in the first embodiment, and "unsupervised learning" is used in the second embodiment.

[0067] The insulation condition diagnostic device 5 operates as the main body of the inference stage. Therefore, it uses the learned model 6 generated by the machine learning device 4 to diagnose (estimate) the insulation condition of the winding 31 based on the measured values ​​measured by the measuring device 3.

[0068] (Machine learning device)

[0069] The machine learning device 4 consists of a general-purpose or special-purpose computer. This computer can be a stationary computer, a portable computer, a client computer, a server computer, or a cloud computer.

[0070] like Figure 3 As shown, the machine learning device 4 includes a control unit 111, an operation unit 112, a display unit 113, a communication unit 114, a media input / output unit 115, a RAM 116, a storage unit 117, and a bus 118. All components of the machine learning device 4 are connected via the bus 118.

[0071] The control unit 111 is composed of a processing unit (CPU, MPU, DSP, etc.), which reads and executes the program loaded into RAM 116, performs various calculations, and thereby controls the operation of each part of the machine learning device 4.

[0072] Furthermore, the control unit 111 may have a plurality of processors, and the plurality of processors may also execute various processes of this embodiment. In this case, the plurality of processors may participate in common processing, or the plurality of processors may independently execute different processes in parallel.

[0073] The operation unit 112 may consist of, for example, a keyboard, mouse, numeric keys, electronic pen, etc., and functions as an input unit.

[0074] The display unit 113 may be composed of, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit.

[0075] The operation unit 112 and the display unit 113 can also be integrated like a touch panel display.

[0076] The communication unit 114 is connected to the measuring device 3 via the network 7 and functions as a communication interface for sending and receiving various data (measurement data, etc.).

[0077] The media input / output unit 115 is composed of, for example, a DVD drive, a CD drive, a USB port, etc., and reads and writes data to media (non-temporary storage media) such as DVD, CD, and USB.

[0078] RAM116 consists of volatile memory (DRAM, SRAM, etc.) that stores various data and programs and functions as main memory.

[0079] The storage unit 117 is composed of storage devices such as HDD and SSD, and stores various data required for the execution of the operating system or program.

[0080] Alternatively, the program can be stored in the storage unit 117 instead of RAM 116. The program can be recorded on a medium as an installable file or an executable file and provided to the machine learning device 4 via the media input / output unit 115. Alternatively, the program can also be provided to the machine learning device 4 by downloading it via the network 7 through the communication unit 114.

[0081] The control unit 111 functions as the learning data acquisition unit 121 and the machine learning unit 122 by executing machine learning programs (not shown).

[0082] The storage unit 117 has a learning data storage unit 123 that functions as a database that stores the learning data acquired by the learning data acquisition unit 121 in multiple groups, and a learned model storage unit 124 that functions as a database that stores the learned learning model 6 that has undergone machine learning.

[0083] Furthermore, in this embodiment, the learning data storage unit 123 and the learned model storage unit 124 are shown as being housed in a single storage device, but they can also be configured as independent storage devices.

[0084] The learning data acquisition unit 121 functions as an interface unit for acquiring learning data, including at least input data, via the communication unit 114, the network 7, or the media input / output unit 115. Alternatively, the learning data acquisition unit 121 can also acquire learning data by receiving input from the operator via the operation unit 112.

[0085] Each sensor (motor voltage and current sensor 101, winding heating state measuring sensor 102, and working fluid measuring sensor 103) serving as a measuring device 3 is installed in each part of the shielded motor pump 2, which functions as a testing machine. This measuring device 3 is connected to the machine learning device 4 via the network 7 to acquire input data contained in the learning data. Furthermore, if the insulation state of the winding at the time of input data acquisition can be determined using image analysis or similar methods, output data contained in the learning data can be automatically acquired in correspondence with the input data. Alternatively, output data contained in the learning data can be acquired by operating the operation unit 112 in correspondence with each input data.

[0086] Furthermore, when learning data is acquired and stored in an external device through other means, the learning data acquisition unit 121 can acquire the learning data through the medium input / output unit 115, or through the communication unit 114 and the network 7.

[0087] In addition, when collecting learning data from the shielded motor pump 2, which serves as a testing machine, it is preferable to use a pump with the same structure as the actual machine used to diagnose the insulation state of the windings. However, if it is confirmed that the measurement data is not dependent on the machine type, a similar shielded motor pump can be used, or a test apparatus that simulates a shielded motor pump can be used.

[0088] The learning data storage unit 123 is a database that stores the learning data acquired by the learning data acquisition unit 121 in multiple groups.

[0089] Therefore, the learning data stored in the learning data storage unit 123 is data loaded with input data measured by the measuring device 3, output data input from the operation unit 112, or existing learning data collected in advance.

[0090] The learned model storage unit 124 is a database that stores the learned models 6 generated by the machine learning unit 122. The learned models 6 stored in the learned model storage unit 124 are provided to the insulation condition diagnostic device 5 via any communication network or storage medium.

[0091] The machine learning unit 122 performs machine learning using learning data stored in the learning data storage unit 123. By sequentially inputting a plurality of sets of learning data into the learning model 6, the machine learning unit 122 enables the learning model 6 to learn the correlation between the input data contained in the learning data and the insulation information of the winding, thereby generating a learned model. There are many methods for generating a learned model, but as a specific method of supervised learning performed by the machine learning unit 122, a neural network can be used, for example.

[0092] Here, the learning data includes at least the voltage and current values ​​supplied to the motor section 20 (motor 24), the temperature of the winding 31 of the stator 23 or related temperatures, temperature differences or pressures, and the inflow temperature and flow rate of the working fluid, as input data. The input data may also include other data, but in this case, it may include data that can be collected without stopping the canned motor pump, i.e., data related to the insulation state of the windings.

[0093] In addition, here is shown an example of using raw data (unprocessed data) as input data, including the voltage and current values ​​supplied to the motor unit 20 (motor 24), the temperature of the winding 31 of the stator 23 or related temperature, temperature difference or pressure data, and the inflow temperature and flow rate of the working fluid. However, this is not a limitation. Data that has undergone preprocessing of at least a portion of the data can also be used as input data (input data acquired by the learning data acquisition unit 121 and input data stored in the learning data storage unit 123 can also be data that has undergone preprocessing of at least a portion of the data).

[0094] Data preprocessing includes, for example, removing outlier values ​​and standardizing the data. It also includes processes commonly performed in machine learning, such as type conversion, handling missing values ​​(removal, interpolation), and scaling. Additionally, preprocessing may include at least one filtering process to ensure the input data meets a specified benchmark.

[0095] In addition, in the case of "supervised learning", the learning data also includes diagnostic information representing the insulation state of the winding when the input data is acquired, such as diagnostic information indicating that the insulation state is a certain state among a plurality of states, as output data (supervised data) corresponding to the input data.

[0096] Here, diagnostic information is information that evaluates the insulation condition of winding 31 (information that evaluates and judges phenomena of insulation degradation), and for example, includes at least one of the following: (1) Determine whether there is information on the working fluid adhering to the winding 31 as a whole (as a working fluid, a large amount of chemical liquid and conductive liquid are used. Therefore, when the housing 33 is broken and the working fluid flows into the stator chamber 45, the working fluid adheres to the winding 31, inducing insulation deterioration). (2) Information for diagnosis based on whether there is burn / discoloration in the condition of winding 31, that is, information for judgment based on whether there is a phenomenon of coil bundle being burned into a ring shape and whether there is a phenomenon of burn in the coil bundle at the entrance (lead opening) of winding 31 (if the winding 31 is damaged, the coil bundle will be heated and burned into a ring shape. In addition, if a large surge voltage is applied to the supplied voltage, the winding 31 will be damaged at the power entrance (lead side), and the coil bundle at the winding entrance will be burned). (3) Information based on whether there is voltage imbalance or whether there is single-phase burnout caused by voltage imbalance (phenomena that may occur due to power supply abnormality). (4) Information judged based on the degree of discoloration of the winding 31 as a whole and whether there are traces of heating (phenomena that may occur due to overheating).

[0097] As a measure of the insulation status of winding 31, the diagnostic information is configured to indicate whether the insulation status is normal or abnormal when diagnosing any abnormalities. In this case, the diagnostic information is defined as "0" when the insulation status of the winding is normal, and as "1" when the insulation status of the winding is abnormal.

[0098] In addition, regarding the insulation abnormality of winding 31, the degree of insulation degradation (degree of abnormality) can also be classified into multiple levels. In this case, the diagnostic information is represented by multiple values: "0" for normal insulation condition, "1" for low-level insulation degradation, and "2" for high-level insulation degradation.

[0099] In addition, abnormalities are not only those that are identified as occurring after the fact during diagnosis, or those that are considered to be within the normal range at the time of diagnosis, but can also include signs of abnormalities that indicate the future occurrence of abnormalities.

[0100] Here, we explain the correlation between the input data (motor voltage and current values, winding temperature or related temperature, temperature difference or pressure, and inflow temperature and flow rate of the working fluid) included in the learning data and the diagnostic information on the insulation status of the windings.

[0101] The voltage and current supplied to the motor unit 20 are applied to the motor 24 from the motor control device 26 when the motor unit 20 is driven to rotate, and therefore change with the output of the motor 24.

[0102] (Correlation between applied voltage and insulation state)

[0103] The applied voltage value is essential information for conducting various cause analyses. For example, when voltage imbalance (three-phase voltage imbalance) increases, current imbalance also increases, leading to burnout in only one phase. Furthermore, measuring the voltage waveform can also identify the presence of high-order harmonic noise or surge voltage, which could be a major cause of burnout.

[0104] Therefore, applying voltage can be used to monitor or predict insulation degradation.

[0105] (Correlation between current value and insulation condition)

[0106] When the motor is driven at a specified rotational speed, the motor current is supplied from the motor control device 26 to the winding 31 of the motor section 20, and therefore varies with the insulation state of the winding 31.

[0107] For example, if the insulation of winding 31 deteriorates excessively, the winding resistance will increase, and the motor will generate more heat. That is, when the insulation resistance increases due to the deterioration of the winding insulation, phenomena such as winding overheating and partial short circuits caused by heat damage will occur.

[0108] Furthermore, if there is a voltage imbalance between the phases of a three-phase power supply, a phenomenon occurs where a larger current flows through only one phase, resulting in only that phase burning out and the winding turning black. Although the relationship between the voltage imbalance rate and the current in each phase varies depending on the number of poles and the output, even a small voltage imbalance rate will confirm a significant difference in the current in each phase.

[0109] Therefore, the motor current value can be used to monitor or predict coil burnout and insulation deterioration.

[0110] (The relationship between winding temperature or related temperature, temperature difference or pressure data and insulation condition)

[0111] The insulation degradation of windings is mainly caused by thermal degradation. That is, because the windings generate heat, the insulation deteriorates, breaks down, and burns. In addition, overheating of the windings can cause the entire winding to burn out and discolor.

[0112] The heat generated by the motor 24 is mainly produced in the winding 31, so it can be said that direct monitoring of the winding is the most preferred method for monitoring the insulation condition of the winding. Moreover, the coil end housing, as a confined space, is the part where heat generated from the coil end tends to accumulate and is difficult to dissipate; therefore, the coil end has the highest temperature in the winding. Thus, the temperature of the winding, especially the temperature of the coil end (T1: the temperature of the coil end away from the pump side, or T2: the temperature of the coil end located on the pump side), or the temperature of the winding 31 wound on the teeth within the stator core 30 (T3), can be used to monitor or predict the insulation condition of the winding.

[0113] Furthermore, since the winding 31 is fixed to the stator core 30, the winding and the stator are directly thermally affected by each other. Additionally, because the heat generated in the winding 31 is transferred as radiant heat to the support sleeves 34a, 34b, end plates 42, 43, bearing cages 44, 46, etc., their surface temperatures vary with the temperature of the winding 31. Furthermore, in this embodiment with internal circulating fluid (e.g., ...), their surface temperatures also vary. Figure 2 As shown, when there is internal circulating fluid flowing inside the housing 33 (as indicated by the arrow), it also changes with the state (amount, temperature) of the internal circulating fluid.

[0114] Therefore, these temperatures related to the winding temperature (T4, T5: temperatures of support sleeves 34a, 34b; T6, T7: temperatures of end plates 42, 43; T8, T9: temperatures of bearing cages 44, 46) can also be used to monitor or predict the insulation status of the winding.

[0115] Furthermore, the pressure P1 in the stator chamber 45 (coil end receiving portions 45a, 45b) increases with the temperature of the winding 31. Moreover, when the winding 31 is burned, organic matter vaporizes, and the pressure in the stator chamber 45 (coil end receiving portions 45a, 45b) easily rises. Therefore, if the correlation accuracy between the temperature of the winding 31 and the pressure in the stator chamber 45 is set within an acceptable range, the pressure in the stator chamber 45 can also be used to monitor or predict the insulation condition of the winding.

[0116] In addition, the pressure of the rotor chamber 50, P2, is measured. By monitoring the difference (ΔP) between the rotor chamber 50 and the stator chamber 45 pressure, P1, damage to the housing 33 (P1 = P2) can be detected. Furthermore, motor malfunctions can be detected.

[0117] Thus, in order to monitor or predict the insulation state of the winding, any one of the following can be used: winding temperature, or temperature related to the winding temperature (e.g., stator core temperature, stator chamber temperature, surface temperature of support sleeves 34a, 34b, surface temperature of end plates 42, 43), temperature difference (e.g., temperature difference between each phase of the winding, temperature difference between each phase of the winding and the working fluid (Δt), temperature difference (T4, T5) between the front and rear support sleeves 34a, 34b, T5 (T5-T4), or pressure related to the winding temperature (e.g., pressure in stator chamber 45), but multiple methods can also be used as needed. In this embodiment, the use of any one of these methods will be described.

[0118] (Correlation between the inflow temperature and flow rate of the working fluid and the insulation condition)

[0119] In the canned motor pump 2, since the housing 33 is in contact with the inner circumferential surface of the stator core 30, the temperature of the stator 23 is mainly affected by the temperature of the working fluid. That is, thermal movement occurs between the stator 23 and the working fluid through the thin, highly thermally conductive housing 33. Therefore, the temperature of the stator 23 is greatly affected by the inflow temperature and flow rate of the working fluid. For example, when the temperature of the working fluid increases, the temperature of the stator core 30 also increases, and the winding 31 further self-heats, thus its temperature is higher than that of the working fluid. Therefore, in the canned motor pump 2, which handles a large amount of high-temperature working fluid, the winding 31 is easily exposed to a high-temperature gas environment where insulation is prone to deterioration.

[0120] Thus, the flow rate and temperature of the working fluid become the basis for evaluating the insulation condition of the winding, and are therefore important for monitoring or predicting the insulation condition.

[0121] The effects of data on motor voltage and current values, temperature or related temperature, temperature difference, or pressure of the winding 31 of the motor section 20, and the inflow temperature and flow rate of the working fluid on the insulation condition are as described above. However, the insulation condition of the winding varies complexly due to the combined effects of these factors. Therefore, it is difficult to properly determine the insulation condition by judging the above data individually. In order to properly determine the insulation condition of the complexly changing winding, it is necessary to consider all the above data comprehensively.

[0122] In addition, such as Figure 2 As shown, even in a shielded motor pump 2 that does not have a cooling device in the motor section 20, the heat generated by the winding 31 in the motor section 20 will be released to the atmosphere through the motor housing 25 (outer cylinder 41, end plates 42 and 43 at both ends, and bearing cage 44), and if the temperature is higher than that of the working fluid, it will also be released to the working fluid through the housing 33.

[0123] However, if the operating conditions of the canned motor pump 2 used in the factory do not change significantly and the heat dissipation conditions are stable, the heat input and output of the canned motor pump can be easily captured. If the data collected under specified conditions and at specified times (data on the voltage and current values ​​supplied to the motor section 20, the temperature of the stator winding or related temperatures, temperature difference or pressure data, and the inflow temperature and flow rate of the working fluid) are used as input data for learning, the tendency of the insulation state of the winding 31 can be grasped.

[0124] In response to this, considering the frequent changes in the voltage and current values ​​supplied to the motor section 20, the temperature of the winding 31 of the stator 23 or related temperatures, temperature differences or pressures, and the inflow temperature and flow rate of the working fluid, as well as the significant changes in heat dissipation to the atmosphere due to atmospheric temperature variations, all of these factors can complexly affect the insulation state of the winding 31. Therefore, in situations where precise input data (input data at a specified time) is required, the accuracy deteriorates. Thus, in such cases, the input data for learning purposes can utilize time-varying data within a specified period.

[0125] That is, data on the voltage and current supplied to the motor section during a specified period, data on the temperature of the stator windings or related temperatures, temperature differences, or pressures during a specified period, and data on the inflow temperature and flow rate of the working fluid during a specified period can also be used as input data. In addition, in the case of "supervised learning," the insulation state of the windings over time during a specified period can also be used as the output data (supervisory data) corresponding to the input data.

[0126] Here, the input and output data within the specified period consist of a plurality of measurement values ​​measured by each sensor at a plurality of measurement times within the specified period. For example, each measurement data is composed of data arranged in chronological order, measured at a specified sampling period within the specified period and associated with the measurement time.

[0127] (Machine learning methods)

[0128] In the case of “supervised learning” as machine learning, the machine learning device 4 forms a learning model, for example, using a neural network model. The neural network model itself is a known model. Input data contained in the learning data is input into the input layer, and the output data output from the output layer as its inference result is compared with the output data (supervised data) contained in the learning data, thereby learning the correlation between the input data and the output data.

[0129] First, as preparation for enabling the machine learning device 4 to perform machine learning, the learning data acquisition unit 121 acquires a desired amount of learning data and stores the acquired multiple sets of learning data in the learning data storage unit 123. The amount of learning data prepared here is appropriately set considering the inference accuracy obtained in the learning model 6.

[0130] Various methods can be used to acquire learning data. For example, in the event of an abnormality in the insulation state of winding 31 or in the event of an abnormality being detected, the measuring device 3 acquires the measured values ​​from each sensor at that moment or within a specified period before or after it. In addition, corresponding to the acquired measured values, a diagnostic result is input from the operation unit 112 (in this case, in the event of an abnormality or the detection of an abnormality, the output data is "1"), thus preparing a set of learning data (input data and output data). The above operation is repeated to prepare multiple sets of learning data.

[0131] Since insulation degradation can sometimes take a long time to occur, and sometimes it can occur suddenly, abnormal insulation states can be intentionally generated to obtain learning data when it is difficult to detect the occurrence or signs of anomalies. In addition, the learning data also includes multiple sets of input and output data for the normal state where the insulation is not abnormal.

[0132] In addition, in order to begin machine learning, a learning model 6 consisting of a neural network model is prepared before learning. In the input layer of this learning model 6, corresponding data are established for the motor voltage and current values, the temperature of the stator winding 31 or related temperatures, temperature differences or pressures, and the inflow temperature and flow rate of the working fluid, which are included in the learning data. In addition, in the output layer, corresponding data are established for the diagnostic information (insulation status) included in the output data of the learning data.

[0133] After the above preparations, Machine Learning Department 122, for example, conducted... Figure 4 The flowchart shown illustrates a series of processes that generate a learned model.

[0134] In step S100, the machine learning unit 122 retrieves a set of learning data from a plurality of sets of learning data stored in the learning data storage unit 123. The learning data can be retrieved in a predetermined order or randomly.

[0135] Then, in step S110, the machine learning unit 122 performs supervised machine learning. That is, the machine learning unit 122 inputs the input data contained in the acquired set of learning data into the prepared learning model, outputs an inference result, compares the inference result with the output data (supervised data) contained in the learning data acquired in step S100, and performs machine learning using a known neural network method. Thus, the machine learning unit 122 enables the learning model 6 to learn the correlation between the input data and the output data (diagnostic information on the insulation state of the winding).

[0136] Then, using the learning model in the learning process, the above-mentioned operation is performed on the remaining learning data to enable the machine learning unit 122 to continue learning. At this time, the machine learning unit 122 determines whether to continue machine learning based on the error between the output data and the supervision data, the number of learning iterations, or the remaining number of unlearned learning data stored in the learning data storage unit 123 (step S120).

[0137] That is, if the machine learning unit 122 determines in step S120 that it will continue machine learning (Yes), it performs the processes of steps S100 to S110 on the learning model 6 that is being learned using the unlearned learning data. If the machine learning unit 122 determines in step S120 that it will not continue machine learning (No), it stores the generated learned learning model 6 in the learned model storage unit 124 in step S130, and ends the machine learning process.

[0138] (Effects of the machine learning device and machine learning method according to the first embodiment)

[0139] According to the machine learning device 4 and machine learning method of the first embodiment described above, a learning model 6 can provide diagnostic information that can accurately infer (diagnose) the insulation state of the winding 31 based on data of voltage and current values ​​supplied to the motor section 20, temperature of the winding 31 of the stator 23 or related temperature, temperature difference or pressure data, and inflow temperature and flow rate data of the working fluid.

[0140] (Insulation condition diagnostic device)

[0141] The insulation state diagnostic device 5, which uses the learned learning model 6 generated by the above method to infer (diagnose) the insulation state of the winding 31, will now be described.

[0142] The insulation condition diagnostic device 5 consists of a general-purpose or dedicated computer. This computer can be a fixed computer, a portable computer, a client computer, a server computer, or a cloud computer.

[0143] In addition, the insulation condition diagnostic device 5 can be installed in the motor control device 26 or in the management device above the motor control device 26.

[0144] Furthermore, the computer constituting the insulation condition diagnostic device 5 may be the same as or different from the computer constituting the machine learning device 4.

[0145] like Figure 5 As shown, the insulation condition diagnostic device 5 includes a control unit 131, an operation unit 132, a display unit 133, a communication unit 134, a media input / output unit 135, a RAM 136, a storage unit 137, and a bus 138. The various parts of the insulation condition diagnostic device 5 are connected via the bus 138.

[0146] The control unit 131 is composed of an arithmetic processing unit (CPU, MPU, DSP, etc.), which reads and executes the program loaded in RAM 136, performs various arithmetic processing, and thereby controls the operation of each part of the insulation condition diagnostic device 5.

[0147] Furthermore, the control unit 131 may have a plurality of processors, and the plurality of processors may also execute various processes of this embodiment. In this case, the plurality of processors may participate in common processing, or the plurality of processors may independently execute different processes in parallel.

[0148] The operation unit 132 may consist of, for example, a keyboard, mouse, numeric keys, electronic pen, etc., and functions as an input unit.

[0149] The display unit 133 may be composed of, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit.

[0150] The operation unit 132 and the display unit 133 can also be integrated like a touch panel display.

[0151] The communication unit 134 is connected to the measuring device 3 via the network 7 and functions as a communication interface for sending and receiving various data (measurement data, etc.).

[0152] The media input / output unit 135 is composed of, for example, a DVD drive, a CD drive, a USB port, etc., and reads and writes data to media (non-temporary storage media) such as DVD, CD, and USB.

[0153] RAM136 consists of volatile memory (DRAM, SRAM, etc.) that stores various data and programs and functions as main memory.

[0154] The storage unit 137 is composed of storage devices such as HDD and SSD, and stores various data required for the execution of the operating system or program.

[0155] Alternatively, the program can be stored in the storage unit 137 instead of RAM 136. The program can be recorded on a medium in the form of an installable file or an executable file and provided to the insulation condition diagnostic device 5 via the media input / output unit 135. Alternatively, the program can also be provided to the insulation condition diagnostic device 5 by downloading it via the network 7 through the communication unit 134.

[0156] The control unit 131 functions as the input data acquisition unit 141, the inference unit 142, and the output processing unit 143 by executing an insulation condition diagnostic program (not shown).

[0157] Additionally, storage unit 137 has a learned model storage unit 144 that functions as a database for storing the learned learning model 6 for machine learning.

[0158] The input data acquisition unit 141 is connected to the measuring device 3 (motor voltage and current sensor 101, winding heating state measuring sensor 102, and working fluid measuring sensor 103) installed on the shielded motor pump 2 (which is the actual machine) via the communication unit 134 and the network 7. It is an interface unit that acquires input data based on the measured values ​​measured by the measuring device 3 (data on motor voltage and current values ​​at a specified time, data on the temperature of the stator winding or related temperature, temperature difference, or pressure, data on the inflow temperature and flow rate of the working fluid, or data on motor voltage and current values ​​during a specified period, data on the temperature of the stator winding or related temperature, temperature difference, or pressure, and data on the inflow temperature and flow rate of the working fluid).

[0159] Additionally, here is shown an example of input data using raw (unprocessed) data of the voltage and current values ​​supplied to the motor unit 20 (motor 24), the temperature of the winding 31 of the stator 23 or related temperature, temperature difference or pressure, and the inflow temperature and flow rate of the working fluid. However, to match the learning data, the input data may also use data that has undergone preprocessing of at least a portion of the voltage and current values ​​supplied to the motor unit 20 (motor 24), the temperature of the winding 31 of the stator 23 or related temperature, temperature difference or pressure, and the inflow temperature and flow rate of the working fluid.

[0160] The inference unit 142 inputs the input data acquired by the input data acquisition unit 141 into the learning model 6 to perform inference processing to infer diagnostic information about the insulation state of the winding 31. In the inference processing, the learned model 6, which has undergone supervised learning and is generated by the machine learning device 4, is used.

[0161] The inference unit 142 not only includes the function of performing inference processing using the learning model 6, but also includes a preprocessing function that, as a preprocessing step, inputs the input data acquired by the input data acquisition unit 141 into the learning model 6 by adjusting the input data into the desired form, etc., as a postprocessing step, and may include a postprocessing function that applies a prescribed logical expression or calculation expression to the value of the output data output from the learning model 6, thereby ultimately diagnosing the insulation state of the winding 31.

[0162] The learned model storage unit 144 is a database storing the learned models 6 used in the inference process of the inference unit 142. Furthermore, the number of learned models 6 stored in the learned model storage unit 144 is not limited to one. For example, multiple learned models 6 can be stored for cases with different input data, different amounts of input data, or different machine learning methods, and can be selected and used appropriately.

[0163] As examples of different input data, for a learning model that uses data on motor voltage and current values, stator winding temperature, and the inflow temperature and flow rate of the aforementioned working fluid, it is equivalent to using data on temperature, temperature difference, or pressure related to the stator winding temperature instead of the stator winding temperature. As for different amounts of input data, it is equivalent to adding data on the left and right heat release to the model of the heat balance of the motor section, etc. (This will be discussed later).

[0164] The output processing unit 143 processes the output inference results from the output inference unit 142, namely, the diagnostic information on the insulation state of the winding 31. Various structures can be used for the specific output mechanism. For example, the output processing unit 143 may notify the operator of the diagnostic information via display or email, or send and store it as a diagnostic record of the shielded motor pump 2 to the motor control device 26 or its superior management device, or use it for drive control of the motor unit 20 or the pump unit 10.

[0165] (Insulation condition diagnosis method)

[0166] Figure 6 An example of an insulation condition diagnosis method using the insulation condition diagnosis device 5 is illustrated using a flowchart. In this example, the diagnostic information for the insulation condition is defined as either "0" for normal conditions or "1" for abnormal conditions.

[0167] First, in step S200, the input data acquisition unit 141 acquires input data based on the measured values ​​measured by the measuring device 3 (motor voltage and current sensor 101, winding heating state measuring sensor 102, working fluid measuring sensor 103) (data on motor voltage and current values ​​at a specified time, data on the temperature of the stator winding or related temperature, temperature difference or pressure, data on the inflow temperature and flow rate of the aforementioned working fluid, or data on motor voltage and current values ​​during a specified period, data on the temperature of the stator winding or related temperature, temperature difference or pressure, and data on the inflow temperature and flow rate of the aforementioned working fluid).

[0168] Then, in step S210, the inference unit 142 inputs the input data into the input layer of the learned learning model 6 to perform inference.

[0169] Then, in step 220, the inference unit 142 determines whether the insulation state is abnormal based on the inference result. That is, as an example of post-processing in supervised learning, the inference unit 142 compares the value of the output data (a number between 0 and 1) with a predetermined threshold to determine whether it is normal or abnormal. For example, if the value of the output data is less than the predetermined threshold, the insulation state is determined to be "no abnormality" (no); if it is above the predetermined threshold, the insulation state is determined to be "abnormal" (yes).

[0170] Then, if the inference result (diagnosis result of the insulation state of the winding) of the inference unit 152 in step 220 is no abnormality (no), the output processing unit 154 outputs information indicating "normal" in step S230. If it is abnormal (yes), the output processing unit 154 outputs information indicating "abnormal" in step 240 (step S240).

[0171] Then, after outputting the diagnostic information in step S230 or S240, the diagnostic process for the insulation status of the winding ends.

[0172] (Effects of the insulation condition diagnostic device and insulation condition diagnostic method of the first embodiment)

[0173] The insulation condition diagnostic device 5 and insulation condition diagnostic method according to the first embodiment described above do not rely on the operator's experience and intuition. In addition, the insulation condition of the winding 31 can be diagnosed with high accuracy without stopping the shielded motor pump 2.

[0174] In other words, even monitoring each input data individually cannot adequately determine the insulation condition under complex and changing conditions. Therefore, in existing technologies, stopping the shielded motor pump, measuring insulation resistance and inter-terminal resistance, and performing trend management largely rely on the operator's experience and intuition. However, according to this diagnostic device, by using a learning model generated from input data such as voltage and current values ​​supplied to the motor section, winding temperature or related temperature, temperature difference or pressure data, and inflow temperature and flow rate data of the working fluid, the insulation condition of the winding can be diagnosed with high accuracy without relying on the experience and intuition of a skilled operator, even when the insulation condition of the winding changes under various influences.

[0175] In the above examples, the heat loss and gain that affect the insulation state of the windings of the shielded motor pump were not actively considered, but the change in heat dissipation cannot be ignored when the motor section 20 has a cooling device.

[0176] Figure 7 An example of a shielded electric motor pump with a cooling device 70 is shown. In this example, a cooling jacket 71 is arranged around the electric motor 20, i.e., on the outside of the outer cylinder 41. The cooling jacket 71 is formed by the outer cylinder 41, an outer peripheral portion 71a disposed on its outer side, and two end plates 71b disposed at both ends of the outer peripheral portion 71a. Cooling pipes 72 are disposed inside the cooling jacket 71, and the space outside the cooling pipes is filled with a heat transfer medium. Here, the heat transfer medium can be, for example, oil, silica, aluminum powder, etc.

[0177] The coolant supplied to the cooling pipe 72 inside the cooling jacket 71 can be either a coolant from a different system than the working fluid of the pump section 10, or it can be the working fluid that flows through the pump section 10. When the working fluid that flows through the pump section 10 is supplied, the working fluid is sent to the rotor chamber 50 after passing through the cooling pipe 72.

[0178] In addition, due to the other structures of the canned motor pump 2 being the same as described above Figure 2 Since the structures shown are the same, the same reference numerals are used for the same parts and the descriptions are omitted.

[0179] With such a cooling device 70 installed, the conditions that serve as the reference for the exposure of the winding 31 will also change depending on the temperature and flow rate of the coolant. Therefore, with such a cooling device 70 installed, a coolant measuring sensor 104 for measuring the inflow temperature and flow rate of the coolant is installed on the pipe 73 that supplies coolant to the cooling jacket 71.

[0180] The coolant measuring sensor 104 is, for example, composed of an inflow temperature sensor 104a and a flow sensor 104b installed on the pipeline supplying coolant to the cooling jacket 71. These inflow temperature sensors 104a and flow sensors 104b can be the same type of sensors as the inflow temperature sensors 103a and flow sensors 103b of the working fluid measuring sensor 103.

[0181] Furthermore, as data measured by the measuring device 3 used in the machine learning device 4 and the insulation condition diagnostic device 5, such as Figure 8A and Figure 8B As shown, the inflow temperature and flow rate of the coolant, measured by the coolant measurement sensor 104, can be added. That is, the inflow temperature and flow rate of the coolant can be included in the input data included in the learning data during the learning phase, and also in the input data for diagnosis during the inference phase.

[0182] In addition, in this case, the coolant inflow temperature and flow rate used as input data are not limited to the raw data (unprocessed data) obtained from the inflow temperature sensor 103a and the flow sensor 103b, but can also be data that has been preprocessed.

[0183] In addition, data on coolant temperature and flow rate can be obtained from data collected at specified times. Furthermore, data collected over a specified period can be used to improve the accuracy of correlation and inference.

[0184] Therefore, when the motor section 20 has a cooling device 70, by further considering the inflow temperature and flow rate of the coolant during the learning and inference phases, the insulation condition of the winding 31 can be inferred (diagnosed) with higher accuracy.

[0185] (Second Implementation)

[0186] In the first embodiment described above, the case of using "supervised learning" as machine learning was explained, but in this embodiment, the case of using "unsupervised learning" is explained. In addition, the basic structure or operation of the measuring device 3, the machine learning device 4 and the insulation condition diagnostic device 5 of the insulation condition diagnostic system 1 of the shielded motor pump 2 or winding 31 constituting the second embodiment are the same as those of the first embodiment, so the description is omitted, and the following description focuses on the differences from the first embodiment.

[0187] Similar to the first embodiment, such as Figure 3 As shown, the machine learning device 4 includes a learning data acquisition unit 121, a learning data storage unit 123, a machine learning unit 122, and a learned model storage unit 124.

[0188] In the case of "unsupervised learning" in machine learning, the learning data only uses diagnostic information to represent the insulation state of the winding at a specified time, or the input data when the insulation state of the winding is in a specified state during a specified period. That is, the learning data does not contain output data (supervisory data).

[0189] As for the insulation status of winding 31, in order to diagnose whether there is an abnormality, the diagnostic information is composed of information indicating whether the insulation status is normal or abnormal. Here, it is composed only of the input data when the insulation status of the winding is normal (data on the voltage and current values ​​supplied to the motor section, the temperature of the stator winding or related temperature, temperature difference or pressure data, the inflow temperature and flow rate of the working fluid, and the inflow temperature and flow rate of the coolant when there is a cooling device).

[0190] When acquiring the aforementioned learning data, the learning data acquisition unit 121 acquires data on the voltage and current values ​​supplied to the motor unit at a predetermined time or period, measured by the measuring device 3 installed on the shielded motor pump 2, which serves as a testing machine; data on the temperature of the stator windings or related temperatures, temperature differences, or pressures; data on the inflow temperature and flow rate of the working fluid; and data on the inflow temperature and flow rate of the coolant (if a cooling device is present). Simultaneously, the operator diagnoses the insulation condition during the timing or predetermined period of measuring the input data, inputting a "normal" result from the operation unit 112. This forms a set of learning data, which is stored in the learning data storage unit 123. This process is repeated to store multiple sets of learning data in the learning data storage unit 123.

[0191] Even in this case, the learning data acquisition unit 121 can collect input data contained in the learning data by installing various sensors (motor voltage and current sensor 101, winding heating state measuring sensor 102, and working fluid measuring sensor 103) as measuring devices 3 in each part of the shielded motor pump 2 which is a test machine, and connecting the measuring devices 3 to the machine learning device 4 via the network 7. If there is already collected learning data (motor voltage and current values ​​as input data, winding temperature or related temperature, temperature difference or pressure data, and working fluid inflow temperature and flow rate data), the learning data acquisition unit 121 can also load the existing learning data into the learning data storage unit 123 via the medium input / output unit 115, the communication unit 114, and the network 7.

[0192] In addition, in this embodiment, the input data is not limited to raw data (unprocessed data), but can also be data that has undergone the same preprocessing as described above on at least a portion of the data.

[0193] The machine learning unit 122 inputs learning data into a learning model, enabling the model to learn the correlation between the input data contained in the learning data and diagnostic information indicating the normal insulation state of the winding, thereby generating a learned model. An autoencoder model can be used, for example, as an unsupervised learning method based on the machine learning unit.

[0194] The autoencoder model itself is a well-known model. It takes input data from the learning data and feeds it into the input layer. By comparing this input data with the output data, which is then output as the inference result from the output layer, it learns the patterns and tendencies of the input data. This series of steps is repeated on the learning model. When the specified learning termination conditions are met, the machine learning ends, and the learned autoencoder model is generated.

[0195] exist Figure 9 The flowchart illustrates an example of a machine learning method using the machine learning apparatus of the second embodiment.

[0196] In this example, as preparation for enabling the machine learning device 4 to perform machine learning, the learning data acquisition unit 121 acquires a desired amount of learning data and stores the acquired multiple sets of learning data in the learning data storage unit 123. The amount of learning data prepared here is appropriately set considering the inference accuracy obtained in the learning model 6.

[0197] Various methods can be used to acquire learning data. For example, by using measuring device 3 to acquire various measured values ​​at a specified time or during a specified period when the insulation state of the winding 31 in the shielded motor pump 2, which serves as a testing machine, is in a normal state, input data constituting learning data can be prepared. Then, by repeatedly performing such operations, multiple sets of learning data can be prepared.

[0198] In addition, in order to begin machine learning, a learning model 6, consisting of an autoencoder model, is prepared before learning. In the input layer of this learning model 6, the voltage and current values ​​supplied to the motor section, the temperature of the stator windings or related temperatures, temperature differences or pressures, and the inflow temperature and flow rate of the working fluid are mapped to the input data, which are included in the learning data. If a cooling device is present, the inflow temperature and flow rate of the coolant are further mapped to the input data.

[0199] After the preliminary preparations are completed, the machine learning unit 122 first retrieves a set of learning data from a plurality of sets of learning data stored in the learning data storage unit 123 in step S300. The learning data can be retrieved in a predetermined order or randomly.

[0200] Then, in step S310, the machine learning unit 122 performs unsupervised machine learning. That is, the machine learning unit 122 inputs the input data contained in the acquired set of learning data into the prepared learning model, outputs the inference result, compares the input data contained in the learning data with the output data output from the output layer as the inference result, and performs machine learning through a known autoencoder method.

[0201] Then, using the learning model in the learning process, the above-mentioned operation is performed on the remaining prepared learning data, so that the machine learning unit 122 continues to learn. At this time, the machine learning unit 122 determines whether to continue machine learning based on the number of learning attempts, the remaining number of unlearned learning data stored in the learning data storage unit 123, etc. (step S320).

[0202] That is, if the machine learning unit 122 determines in step S320 that it will continue machine learning (Yes), it performs the processes of steps S300 to S310 on the learning model 6 that is being learned using the unlearned learning data. If, in step S320, it determines that the machine learning unit 122 will not continue machine learning (No), then in step S330, the machine learning unit 122 stores the generated learned model 6 in the learned model storage unit 124 and ends the machine learning process.

[0203] (Effects of the machine learning device and machine learning method according to the second embodiment)

[0204] If the machine learning device and machine learning method of the second embodiment above are used, a learning model 6 can be provided that can accurately infer (estimate) diagnostic information of the insulation state of the winding based on data of motor voltage and current values ​​at a specified time, temperature of stator windings or related temperature, temperature difference or pressure, and inflow temperature and flow rate of working fluid, or data of motor voltage and current values ​​during a specified period, temperature of stator windings or related temperature, temperature difference or pressure, and inflow temperature and flow rate of the aforementioned working fluid.

[0205] In this embodiment, the insulation condition diagnostic device 5 is in Figure 5 The insulation condition diagnostic device 5 has the same structure as the first embodiment. That is, the insulation condition diagnostic device 5 has an input data acquisition unit 141, an inference unit 142, an output processing unit 143, and a learned model storage unit 144.

[0206] Similar to the first embodiment, the inference unit 152 inputs the input data acquired by the input data acquisition unit 141 into the learned learning model 6 to perform inference processing to infer diagnostic information of the insulation state of the winding. However, in the inference processing, the learned model that has undergone unsupervised learning in the machine learning device 4 is used.

[0207] Figure 10 This is a flowchart illustrating an example of an insulation condition diagnosis method for the winding 31 of the insulation condition diagnosis device 5 based on the second embodiment. In this example, the insulation condition diagnosis information is explained as being defined as either "0" when normal and "1" when abnormal.

[0208] First, in step S400, the input data acquisition unit 141 acquires input data based on the measured values ​​measured by the measuring device 3 (motor voltage and current values ​​at a specified time or period, temperature of the stator winding or related temperature, temperature difference or pressure, inflow temperature and flow rate of the working fluid, and in the case of a cooling device 70, inflow temperature and flow rate of the coolant).

[0209] Here, the input data can also be data that has undergone preprocessing of a portion of the data.

[0210] Then, in step S410, the inference unit 142 inputs input data to the input layer of the learning model 6 to perform inference and obtains output data output from the output layer of the learning model 6.

[0211] Then, in step S420, based on the inference result, it is determined whether the insulation state is abnormal. That is, as an example of post-processing for unsupervised learning, the inference unit 142 calculates the difference between the feature quantity based on the input data and the feature quantity based on the output data. If the difference is less than a predetermined threshold, the insulation state is determined to be "no abnormality" (No); if it is above the predetermined threshold, the insulation state is determined to be "abnormal" (Yes).

[0212] Next, in step S430, if the output processing unit 154 determines that the diagnostic information of the insulation state of the winding, which is the inference result of the inference unit 152, is no abnormal (no), it outputs information indicating "normal" (step S430); if it determines that it is "abnormal" (yes), it outputs information indicating "abnormal" (step S440).

[0213] Then, after outputting the diagnostic information in step S430 or S440, the diagnostic process for the insulation status of the winding ends.

[0214] (Effects of the insulation condition diagnostic device and insulation condition diagnostic method according to the second embodiment)

[0215] In the insulation condition diagnosis device 5 and insulation condition diagnosis method of the second embodiment described above, the insulation condition of the winding can also be diagnosed with high accuracy without relying on the operator's experience and intuition, and without stopping the shielded motor pump 2.

[0216] (Other implementation methods)

[0217] Furthermore, in the above embodiments, specific machine learning methods for the machine learning unit 122 were described using neural networks (first embodiment) and autoencoders (second embodiment), respectively. However, the machine learning unit 122 can also employ any other arbitrary machine learning methods. For example, it can also employ tree-based methods such as decision trees and regression trees, ensemble learning such as bagging and boosting, neural network-based methods such as recurrent neural networks and convolutional neural networks (including deep learning), hierarchical clustering, non-hierarchical clustering, clustering methods such as k-nearest neighbors and k-means, multivariate analysis such as principal component analysis, factor analysis, logistic regression, support vector machines, etc.

[0218] In addition, the aforementioned machine learning device 4 and insulation condition diagnostic device 5 may also be provided as programs (machine learning program and insulation condition diagnostic program) for performing the various steps of the aforementioned machine learning method and insulation condition diagnostic method.

[0219] In the above example, the temperature of the stator 23 winding 31, or related temperature, temperature difference, or pressure data, as well as measurement data from each sensor at specified times or during specified periods, are used as input data for learning purposes. However, the input data can also appropriately use temperature difference data related to the temperature of the winding 31. For example, even if the voltage and current values ​​supplied to the motor section 20 are constant, if the winding resistance values ​​of each phase become unbalanced due to insulation deterioration, the temperature of each phase winding will also deviate. Therefore, the temperature difference data of each phase winding can also be used.

[0220] Furthermore, as temperature difference data related to the temperature of winding 31, even if the temperature difference between each phase of the winding and the working fluid (Δt), and the temperature difference (T5-T4) between the front and rear support sleeves 34a and 34b, abnormal heating of the motor (winding 31) can be detected, these temperature difference data can also be used as input data.

[0221] (Embodiments of the present invention)

[0222] Hereinafter, for embodiments of the present invention mastered according to the above-described embodiments, the terms and symbols described in each embodiment will be referenced and described below.

[0223] A first embodiment of the present invention is a machine learning device (e.g., machine learning device 4) that generates a learning model (e.g., learning model 6) for an insulation condition diagnostic device (e.g., insulation condition diagnostic device 5), the insulation condition diagnostic device being used to diagnose the insulation condition of the stator windings (e.g., winding 31) used in a canned motor pump (e.g., canned motor pump 2). The shielded electric motor pump has an electric motor section (e.g., electric motor section 20) and a pump section (e.g., pump section 10) driven by the electric motor section. The motor unit includes: a rotor (e.g., rotor 22) rotatable about a shaft (e.g., shaft 21); a stator (e.g., stator 23) opposite the rotor with a gap; and a cylindrical housing (e.g., housing 33) forming a rotor chamber (e.g., rotor chamber 50) on the inside to house the rotor, and a stator chamber (e.g., stator chamber 45) on the outside together with the motor housing (e.g., motor housing 25) to house the stator. The shielded motor pump circulates a portion of the working fluid in the pump section to the rotor chamber, wherein... The machine learning device has: The learning data acquisition unit (e.g., learning data acquisition unit 121) takes unprocessed data or data that has undergone preprocessing on at least a portion of the unprocessed data as input data, and acquires a plurality of sets of learning data that at least contain the input data. The unprocessed data includes data on voltage and current values ​​supplied to the motor unit, the temperature of the stator windings (e.g., windings 31, coil ends 31a, 31b) and related temperatures (e.g., stator core 30, support sleeves 34a, 34b, end plates 42, 43, or bearing cages 44, 46), temperature differences (e.g., temperature differences between each phase of the windings, the temperature difference between each phase of the windings and the working fluid (Δt), the temperature difference (T4, T5) between the front and rear support sleeves 34a, 34b (T5-T4)), or pressure (e.g., pressure of the stator chamber 45 or the coil end receiving portions 45a, 45b), and the inflow temperature and flow rate of the working fluid. The machine learning unit (e.g., machine learning unit 122) learns the correlation between the input data and diagnostic information about the insulation state of the winding by inputting the learning data into the learning model; and The learned model storage unit (e.g., the learned model storage unit 124) stores the learned model learned by the machine learning unit.

[0224] The input data used as learning data is either unprocessed data or data for which at least a portion has been preprocessed, as described in (1) to (3) below.

[0225] (1) Data on the voltage and current values ​​supplied to the motor unit, (2) Data on the temperature of the stator windings or related temperature, temperature difference, or pressure. (3) Data on the inflow temperature and flow rate of the working fluid.

[0226] (1) to (3) are the required input data, but (2) can be any of the following: (i) Temperature data of the stator windings, (ii) The temperature of the stator windings and related temperature data. (iii) Data on the temperature of the stator windings and related temperature differences. (iv) Data on the temperature and related pressure of the stator windings.

[0227] According to this structure, a learning model can be provided that can accurately infer diagnostic information about the winding insulation of the motor section based on data such as voltage and current supplied to the motor section of the canned motor pump, temperature or related temperature, temperature difference or pressure data of the stator winding of the motor section, and inflow temperature and flow rate data of the working fluid in the pump section—either unprocessed data or data preprocessed on at least a portion of the unprocessed data. Furthermore, a learning model can be provided that can accurately diagnose the insulation condition of the windings without stopping the canned motor pump.

[0228] The second embodiment of the present invention is a machine learning device. In the first embodiment, at least a portion of the motor housing of the motor part is further provided with a cooling jacket (e.g., cooling jacket 71). The input data further includes data on the inflow temperature and flow rate of the coolant supplied into the cooling jacket, i.e., unprocessed data or data on at least a portion of the unprocessed data that has been preprocessed.

[0229] According to this structure, as input data included in the learning data, based on the input data of the first embodiment, data on the inflow temperature and flow rate of the coolant are further added, namely unprocessed data or data on at least a portion of the unprocessed data that has been preprocessed. Therefore, a learning model that can accurately infer (estimate) diagnostic information on the winding insulation of the motor section can be provided.

[0230] The third embodiment of the present invention is a machine learning device. In the first embodiment, the voltage and current data supplied to the motor section are data supplied to the motor section during a specified period, the temperature of the stator winding or the temperature, temperature difference or pressure related thereto are data during the specified period, and the inflow temperature and flow rate of the working fluid are data during the specified period.

[0231] According to this structure, data from a specified period is used as input data in the first embodiment, thus improving the accuracy of the correlation between the data being learned.

[0232] The fourth embodiment of the present invention is a machine learning device. In the second embodiment, the voltage and current data supplied to the motor are data supplied to the motor during a specified period, the temperature of the stator winding or related temperature, temperature difference or pressure data are data during the specified period, the inflow temperature and flow rate of the working fluid are data during the specified period, and the inflow temperature and flow rate of the coolant supplied to the cooling jacket are data during the specified period.

[0233] According to this structure, data from a specified period is used as input data in the second embodiment, thus improving the accuracy of the correlation between the learning data.

[0234] The fifth embodiment of the present invention is a machine learning device. In any one of the first to fourth embodiments, the learning data further includes diagnostic information corresponding to the input data, indicating which of a plurality of states the insulation state of the winding is as output data. The machine learning unit enables the learning model to learn the correlation between the input data and the output data through supervised learning.

[0235] Based on this structure, a learning model can be generated that learns the correlation between input data and diagnostic information about the insulation state of the winding through supervised learning.

[0236] The sixth embodiment of the present invention is a machine learning device. In any one of the first to fourth embodiments, the learning data only includes input data when the diagnostic information indicates that the insulation state of the winding is in a specified state. The machine learning unit enables the learning model to learn the correlation between the input data and the diagnostic information indicating that the insulation state of the winding is in the specified state through unsupervised learning.

[0237] Based on this structure, a learning model can be generated that learns the correlation between input data and diagnostic information about the insulation state of the winding through unsupervised learning.

[0238] A seventh embodiment of the present invention is an insulation condition diagnostic device (e.g., insulation condition diagnostic device 5) for diagnosing the insulation condition of the stator windings used in the shielded motor pump using a learning model generated by the machine learning device of any one of the embodiments from the first to the sixth. The insulation condition diagnostic device has the following features: An input data acquisition unit (e.g., input data acquisition unit 141) acquires input data that includes unprocessed data or data for which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor unit, temperature or related temperature, temperature difference, or pressure data of the stator windings, inflow temperature and flow rate of the working fluid, and inflow temperature and flow rate data of coolant, which can be selected as needed. The inference unit (e.g., inference unit 142) inputs the input data acquired by the input data acquisition unit into the learning model to infer diagnostic information about the insulation state of the stator winding.

[0239] According to this structure, diagnostic information on the insulation condition of the stator winding can be inferred using input data that includes data on the voltage and current values ​​supplied to the motor section, the temperature of the stator winding or related temperature, temperature difference or pressure data, the inflow temperature and flow rate of the working fluid, or the inflow temperature and flow rate of the coolant that can be selected as needed, i.e., unprocessed data or data on at least a portion of the unprocessed data. Therefore, the insulation condition of the winding can be diagnosed with high accuracy without stopping the factory.

[0240] The eighth embodiment of the present invention is a machine learning method for learning a learning model, which is used in an insulation condition diagnostic device to diagnose the insulation condition of the stator windings used in a shielded motor pump. The shielded electric motor pump has an electric motor section and a pump section driven by the electric motor section. The motor unit includes: a rotor capable of rotating about an axis; a stator opposite the rotor with a gap; and a cylindrical housing having an inner rotor chamber for accommodating the rotor and an outer stator chamber for accommodating the stator together with the motor housing. The shielded motor pump circulates a portion of the working fluid in the pump section to the rotor chamber, wherein... The machine learning method has the following characteristics: The learning data acquisition process (e.g., processing S100, S300) takes unprocessed data or data that has undergone preprocessing on at least a portion of the unprocessed data as input data, and acquires a plurality of sets of learning data that at least have the input data. The unprocessed data includes data on voltage and current values ​​supplied to the motor section, data on the temperature of the stator windings or related temperatures, temperature differences or pressures, and data on the inflow temperature and flow rate of the working fluid. The machine learning process (e.g., processing S110, S310) involves inputting the learning data into the learning model, thereby enabling the learning model to learn the correlation between the input data and diagnostic information regarding the insulation state of the winding; and The learned model storage process (e.g., processing S130, S330) stores the learned model learned by the machine learning process into the learned model storage unit.

[0241] According to this structure, a learning model can be provided that can accurately infer (estimate) diagnostic information about the winding insulation of the motor section based on data such as voltage and current values ​​supplied to the motor section of the canned motor pump, temperature or related temperature, temperature difference or pressure data of the stator windings of the motor section, and inflow temperature and flow rate data of the working fluid in the pump section—either unprocessed data or data preprocessed on at least a portion of the unprocessed data. Furthermore, a learning model can be provided that can accurately diagnose the insulation condition of the windings without stopping the canned motor pump.

[0242] The ninth embodiment of the present invention is a machine learning method. In the eighth embodiment, the motor portion further includes a cooling jacket on at least a portion of the outer periphery of the motor housing. The input data also includes data on the inflow temperature and flow rate of the coolant supplied to the cooling jacket, i.e., unprocessed data or data on at least a portion of the unprocessed data that has undergone preprocessing.

[0243] According to this structure, as input data included in the learning data, based on the input data of the eighth embodiment, data on the inflow temperature and flow rate of the coolant are further added, namely, unprocessed data or data on at least a portion of the unprocessed data that has been preprocessed. Therefore, a learning model that can accurately infer (estimate) diagnostic information on the winding insulation of the motor section can be provided.

[0244] The tenth embodiment of the present invention is a machine learning program for causing a computer to execute the various steps of the machine learning method described in the eighth or ninth embodiment.

[0245] According to this structure, a learning model can be provided that can accurately infer (estimate) diagnostic information about the winding insulation of the motor section based on data such as voltage and current supplied to the motor section of the canned motor pump, temperature or related temperature, temperature difference or pressure of the stator winding of the motor section, inflow temperature and flow rate of the working fluid in the pump section, and inflow temperature and flow rate of coolant (selectable as needed)—either unprocessed data or data preprocessed on at least a portion of the unprocessed data. Furthermore, a learning model can be provided that can accurately diagnose the insulation condition of the windings without stopping the canned motor pump.

[0246] The eleventh embodiment of the present invention is an insulation condition diagnosis method, which uses a learning model generated by the machine learning device described in any one of the first to sixth embodiments to diagnose the insulation condition of the stator windings used in the shielded motor pump. The insulation condition diagnosis method has the following characteristics: The input data acquisition process (e.g., processing S200, S400) acquires input data containing unprocessed data or data for which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor section, temperature or related temperature, temperature difference, or pressure data of the stator windings, inflow temperature and flow rate of the working fluid, and inflow temperature and flow rate data of the coolant, which can be selected as needed. The inference process (processing S210, S410) inputs the input data obtained by the input data acquisition process into the learning model to infer the diagnostic information of the insulation state of the stator winding.

[0247] According to this structure, it is possible to accurately infer (estimate) diagnostic information about the winding insulation of the motor section using input data that includes data on the voltage and current values ​​supplied to the motor section, the temperature of the stator windings or related temperatures, temperature differences or pressures, the inflow temperature and flow rate of the working fluid, and the inflow temperature and flow rate of the coolant, which can be selected as needed. This input data includes unprocessed data or data on which at least a portion of the unprocessed data has been preprocessed. Furthermore, the insulation condition of the windings can be diagnosed with high accuracy without stopping the canned motor pump.

[0248] The twelfth embodiment of the present invention is an insulation condition diagnosis program for causing a computer to execute the various steps of the insulation condition diagnosis method described in the eleventh embodiment.

[0249] According to this structure, diagnostic information regarding the winding insulation of the motor section can be accurately inferred (estimated) based on input data including voltage and current values ​​supplied to the motor section of the canned motor pump, temperature or related temperature, temperature difference, or pressure data of the stator windings of the motor section, inflow temperature and flow rate data of the working fluid in the pump section, and inflow temperature and flow rate data of coolant that can be selected as needed—either unprocessed data or data input data for which at least a portion of the unprocessed data has been preprocessed. Furthermore, the insulation condition of the windings can be diagnosed with high accuracy without stopping the canned motor pump.

[0250] Explanation of reference numerals in the attached figures: 1 Insulation Condition Diagnostic System 2. Shielded motor pump 3. Measuring apparatus 4 Machine Learning Devices 5. Insulation condition diagnostic device 6 Learning Model 10. Pump Section 20 Electric Motor Section 21 axis 22 Rotors 23 Stator 25 Motor housing 31 windings 33. Shell 45 Stator Chamber 50 Rotor Chamber 71 Cooling jacket 101 Motor voltage and current sensor 102 Winding Heating Status Measurement Sensor 103 Working Fluid Measurement Sensor 104 Coolant Measurement Sensor 121 Learning Data Acquisition Department 122 Machine Learning Department 123 Learning Data Storage Department 124 Learned Model Storage Department 141 Input Data Acquisition Department 142 Inference Department 143 Output Processing Unit 144 Learned Model Storage Department

Claims

1. A machine learning apparatus for generating a learning model for an insulation condition diagnostic device used to diagnose the insulation condition of stator windings in a shielded motor pump. The shielded electric motor pump has an electric motor section and a pump section driven by the electric motor section. The motor unit includes: a rotor capable of rotating about an axis; a stator opposite the rotor with a gap; and a cylindrical housing having an inner rotor chamber for accommodating the rotor and an outer stator chamber for accommodating the stator together with the motor housing. The shielded motor pump circulates a portion of the working fluid in the pump section to the rotor chamber, wherein... The machine learning device has: The learning data acquisition unit takes unprocessed data or data that has undergone preprocessing on at least a portion of the unprocessed data as input data, and acquires a plurality of sets of learning data having at least the input data. The unprocessed data includes data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator windings or related temperatures, temperature differences or pressures, and data on the inflow temperature and flow rate of the working fluid. The machine learning unit, by inputting the learning data into the learning model, enables the learning model to learn the correlation between the input data and diagnostic information regarding the insulation state of the winding; and The learned model storage unit stores the learned models learned by the machine learning unit.

2. The machine learning apparatus according to claim 1, wherein, The motor unit also has a cooling jacket on at least a portion of the motor housing. The input data also includes data on the inflow temperature and flow rate of the coolant supplied into the cooling jacket.

3. The machine learning apparatus according to claim 1, wherein, The voltage and current values ​​supplied to the motor unit are data supplied to the motor unit within a specified period. The data on the temperature of the stator windings, or related temperature, temperature difference, or pressure, are data from the specified period. The data on the inflow temperature and flow rate of the working fluid are data from the specified period.

4. The machine learning apparatus according to claim 2, wherein, The voltage and current values ​​supplied to the motor unit are data supplied to the motor unit within a specified period. The data on the temperature of the stator windings, or related temperature, temperature difference, or pressure, are data from the specified period. The data on the inflow temperature and flow rate of the working fluid are data from the specified period. The data on the inflow temperature and flow rate of the coolant supplied to the cooling jacket are data within the specified period.

5. The machine learning apparatus according to any one of claims 1 to 4, wherein, The learning data also includes diagnostic information corresponding to the input data, indicating which of a plurality of states the insulation state of the winding is in, as output data. The machine learning unit enables the learning model to learn the correlation between the input data and the output data through supervised learning.

6. The machine learning apparatus according to any one of claims 1 to 4, wherein, The learning data only includes input data when the diagnostic information indicates that the insulation state of the winding is in a specified state. The machine learning unit enables the learning model to learn the correlation between the input data and the diagnostic information representing the insulation state of the winding as the specified state through unsupervised learning.

7. An insulation condition diagnostic device for diagnosing the insulation condition of the stator windings used in the shielded motor pump using a learning model generated by the machine learning device of claim 1 or 3, wherein, The insulation condition diagnostic device has the following features: The input data acquisition unit acquires input data that includes unprocessed data or data for which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor unit, temperature of the stator windings or related temperatures, temperature difference or pressure data, and inflow temperature and flow rate data of the working fluid. as well as The inference unit inputs the input data acquired by the input data acquisition unit into the learning model to infer diagnostic information about the insulation state of the stator winding.

8. An insulation condition diagnostic device for diagnosing the insulation condition of the stator windings used in the shielded motor pump using a learning model generated by the machine learning device of claim 2 or 4, wherein, The insulation condition diagnostic device has the following features: The input data acquisition unit acquires input data that includes unprocessed data or data for which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor unit, temperature of the stator windings or related temperatures, temperature differences or pressures, inflow temperature and flow rate of the working fluid, and inflow temperature and flow rate of the coolant. as well as The inference unit inputs the input data acquired by the input data acquisition unit into the learning model to infer diagnostic information about the insulation state of the stator windings.

9. A machine learning method for learning a learning model, said learning model being used in an insulation condition diagnostic device for diagnosing the insulation condition of stator windings used in a shielded motor pump. The shielded electric motor pump has an electric motor section and a pump section driven by the electric motor section. The motor unit includes: a rotor capable of rotating about an axis; a stator opposite the rotor with a gap; and a cylindrical housing having an inner rotor chamber for accommodating the rotor and an outer stator chamber for accommodating the stator together with the motor housing. The shielded motor pump circulates a portion of the working fluid in the pump section to the rotor chamber, wherein... The machine learning method has the following characteristics: The learning data acquisition process takes unprocessed data or data in which at least a portion of the unprocessed data has been preprocessed as input data, and acquires a plurality of sets of learning data having at least the input data. The unprocessed data includes data on voltage and current values ​​supplied to the motor section, data on the temperature of the stator windings or related temperatures, temperature differences or pressures, and data on the inflow temperature and flow rate of the working fluid. The machine learning process involves inputting the learning data into the learning model, thereby enabling the learning model to learn the correlation between the input data and diagnostic information regarding the insulation state of the winding; and The learned model storage process stores the learned model learned by the machine learning process into the learned model storage unit.

10. The machine learning method according to claim 9, wherein, The motor unit further includes a cooling jacket on at least a portion of the outer periphery of the motor housing. The input data also includes data on the inflow temperature and flow rate of the coolant supplied into the cooling jacket.

11. A machine learning program, wherein, The machine learning program is used to enable a computer to perform the various steps of the machine learning method of claim 9 or 10.

12. An insulation condition diagnosis method, comprising using a learning model generated by the machine learning apparatus of claim 1 or 3 to diagnose the insulation condition of the stator windings used in the shielded motor pump, wherein, The insulation condition diagnosis method has the following characteristics: The input data acquisition process acquires input data that includes unprocessed data or data in which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor unit, temperature of the stator windings or related temperatures, temperature differences or pressures, and inflow temperature and flow rate of the working fluid. as well as The inference process involves inputting the input data obtained from the input data acquisition process into the learning model to infer diagnostic information about the insulation state of the stator windings.

13. An insulation condition diagnosis method, comprising using a learning model generated by the machine learning apparatus of claim 2 or 4 to diagnose the insulation condition of the stator windings used in the shielded motor pump, wherein, The insulation condition diagnosis method has the following characteristics: The input data acquisition process acquires input data that includes unprocessed data or data for which at least a portion of the unprocessed data has been preprocessed. The unprocessed data includes voltage and current values ​​supplied to the motor unit, temperature of the stator windings or related temperatures, temperature differences or pressures, inflow temperature and flow rate of the working fluid, and inflow temperature and flow rate of the coolant. as well as The inference process involves inputting the input data obtained from the input data acquisition process into the learning model to infer diagnostic information about the insulation state of the stator windings.

14. An insulation condition diagnostic procedure, wherein, The insulation condition diagnostic program causes the computer to execute the steps of the insulation condition diagnostic method of claim 12.

15. An insulation condition diagnostic procedure, wherein, The insulation condition diagnostic program causes the computer to execute the steps of the insulation condition diagnostic method of claim 13.

Images